Hematopoietic stem cell niche generation and maintenance are distinguishable by an epitranscriptomic program

Summary

The niche is typically considered as a pre-established structure sustaining stem cells. Therefore, the regulation of its formation remains largely unexplored. Whether distinct molecular mechanisms control the establishment versus maintenance of a stem cell niche is unknown. To address this, we compared perinatal with adult bone marrow mesenchymal stromal cells (MSCs), a key component of the hematopoietic stem cell (HSC) niche. MSCs exhibited enrichment in genes mediating m⁶A mRNA methylation at the perinatal stage and downregulated the expression of Mettl3, the m⁶A methyltransferase, shortly after birth. Deletion of Mettl3 from developing MSCs, but not osteoblasts, led to excessive osteogenic differentiation and a severe HSC niche formation defect, which was significantly rescued by deletion of Klf2, an m⁶A target. In contrast, deletion of Mettl3 from MSCs postnatally did not affect HSC niche. Stem cell niche generation and maintenance thus depend on divergent molecular mechanisms, which may be exploited for regenerative medicine.

Graphical Abstract

eTOC Blurb

The molecular mechanisms regulating stem cell niche formation are poorly understood. Here an epitranscriptomic program relying on Mettl3 is uncovered that is distinctly important for hematopoietic stem cell niche formation but not maintenance.

Introduction

Interaction between stem cells and their niche is critical for growth, homeostasis, and tissue repair. Although stem cells dynamically change properties throughout life, the niche has often been considered as a static preexisting structure that supports stem cells under homeostasis. As a result, little attention has been paid to the molecular mechanisms that regulate the generation of a stem cell niche.

The dynamic behavior of stem cells within their niches has been well characterized in the hematopoietic system. Hematopoietic stem cells (HSCs) emerge from hemogenic endothelium in the aorta-gonads-mesonephros (AGM) region at embryonic day (E) 9.5–11 in mice. Thereafter, HSCs change their intrinsic properties to adapt to extrinsic environments and physiological needs ^1–4. Starting around E11.5, HSCs seed and expand in the fetal liver, where hepatic stellate and endothelial cells are major niche components, elaborating key cytokines, such as SCF ⁵. Around birth, HSCs further migrate into the newly-generated bone marrow niche and reside there throughout adult life ⁶. Within the nascent bone marrow niche, fetal HSCs change their cell cycle status, function, and transcriptome by 2–4 weeks after birth to an adult phenotype ^3,7.

Much has been learned about the adult bone marrow HSC niche, where leptin receptor-expressing (LepR⁺) mesenchymal stromal cells (MSCs) serve as a critical component by secreting HSC maintenance factors, including SCF, CXCL12, and pleiotrophin ^8–12. They are also the major source of IL7, a critical factor for lymphoid cell development ¹³. Although HSCs can be found in the embryonic bone marrow, LepR⁺ cells start to emerge in the neonatal bone marrow after birth ^14–16, coinciding with and likely allowing the lodging of HSCs in the newly-formed bone marrow niche. Perichondrial MSCs expressing Osx, Acan, Col2, Sox9, Hoxa11, PthrP, and Gli1 give rise to LepR⁺ cells during ontogeny ^17–23. These LepR⁺ stromal cells are responsible for the vast majority of the in vitro CFU-F activity in the adult bone marrow and can differentiate to adipocytes and osteoblasts in adults in vivo ¹⁴. Beyond the local niche, HSCs are also under the regulation of distant organ-derived systemic factors, such as TPO and estrogen ^24–26.

The identification of MSCs as a key component of the adult bone marrow HSC niche has spurred the search for molecular mechanisms that regulate the niche function. However, most studies have focused mainly on the mechanisms that maintain the adult bone marrow niche. For example, transcriptional factors and epigenetic regulators, including Foxc1, Bmi1, Ebf3, Ebf1, and Runx1, are preferentially expressed by adult bone marrow MSCs compared with other stromal cells ^27–31. Consistently, deletion of these factors from bone marrow MSCs leads to defects in the function of the adult HSC niche, suggesting that these genes are required for HSC niche maintenance in the bone marrow. Despite these progresses, the molecular mechanisms that regulate niche generation in the bone marrow remain unclear. No gene is known to specifically control the generation but not maintenance of the bone marrow HSC niche.

N⁶-methyladenosine (m⁶A) is the most abundant internal chemical modification on mammalian mRNAs and an emerging novel molecular mechanism that plays important and wide-ranging roles in regulating gene expression at multiple levels in diverse biological processes. A writer protein complex composed of multiple subunits, including the sole critical catalytic component METTL3 and the related protein METTL14, deposits m⁶A onto target mRNAs and regulates gene expression at the epitranscriptomic level through readers, such as YTH and Igf2bp proteins ^32,33. In this study, we identified Mettl3-mediated m⁶A mRNA modification in MSCs as a specific molecular mechanism for the generation of the bone marrow HSC niche.

Results

Genes associated with m⁶A modification are preferentially expressed by developing bone marrow MSCs

Recent single-cell RNA sequencing (scRNA-seq) profiling efforts have revealed the cellular components of the bone marrow niche during ontogeny ^15,16. To identify molecular mechanisms that specifically regulate the generation of the bone marrow HSC niche during ontogeny, we compared bone marrow stroma between perinatal stage (including E16.5, E18.5 and postnatal day (P) 0), when the bone marrow niche is forming, and adulthood, when the bone marrow niche is maintained at steady state, by mining a public scRNA-seq dataset ¹⁵. We focused on MSCs as they are a key component of the bone marrow HSC niche ^8–12. Our analysis identified an MSC cluster that expressed the highest levels of HSC niche factors, including Scf and Cxcl12 (Figures S1A, 1A and 1B).

Figure 1. Genes mediating m6A modification are preferentially expressed by developing MSCs in the bone marrow niche. See also Figure S1.

(A) Uniform manifold approximation and projection (UMAP) visualization showing cell clusters in bone marrow stroma identified by scRNA-seq. Data from perinatal and adult mice are shown. Circles highlight MSCs (cluster 9).

(B) Violin plots showing the expression levels of MSC markers in bone marrow stroma.

(C) Violin plots showing the expression levels of indicated genes in perinatal and adult bone marrow MSCs.

(D) GSEA revealed that mRNA processing, RNA methylation, methyltransferase complex, and mRNA m⁶A modification gene sets are significantly enriched in bone marrow MSCs at perinatal compared with adult stages.

(E) Violin plots comparing the expression levels of genes involved in the m⁶A pathway in perinatal and adult bone marrow MSCs.

(F) RT-qPCR data showing progressive downregulation of Mettl3 in bone marrow MSCs within two weeks after birth.

(G and H) Flow cytometry comparing METTL3 protein levels in CD45/Ter119⁻PDGFRα⁺ bone marrow MSCs from P2 and adult mice. MFI, median fluorescent intensity.

(I) ELISA comparing mRNA m⁶A modification abundance in cultured P2 and adult bone marrow stromal cells.

Data were presented as mean ± SD. P values were calculated with two-tailed unpaired t-test in C, E, H, and I and with one-way ANOVA followed by Dunnett’s multiple comparisons test in F.

Bone marrow MSCs expressed considerable levels of Pdgfra and Prx1 at both perinatal and adult stages (Figure 1C). Compared with their perinatal counterparts, adult bone marrow MSCs expressed higher levels of Lepr, Cxcl12, Scf, and Pdgfrb as well as known regulators for the maintenance of HSC niche, including Foxc1, Ebf1/3, Runx1/2, and Bmi1 (Figure 1C). In line with a prior report ¹⁶, perinatal bone marrow MSCs preferentially expressed Col3a1, Ptn, and Aspn (Figure 1C). To systematically explore the pathways that are specifically active in perinatal MSCs, we performed gene set enrichment analysis (GSEA) using all the gene sets from gene ontology (GO) molecular function (MF) ontology. We found that gene sets involving cytokine receptor activity were enriched in adult MSCs, while gene sets related to mRNA binding were among the most enriched in perinatal MSCs (Figures S1B–S1D). Detailed analysis of the core enrichment genes in mRNA binding signature revealed that m⁶A-containing RNA binding activity was among the most significant enriched terms in perinatal MSCs (Figure S1E). Consistently, gene sets involved in mRNA processing, RNA methylation, and methyltransferase complex, particularly mRNA m⁶A modification, were enriched in MSCs at perinatal compared with adult stages (Figure 1D), prompting us to further examine genes involved in mRNA m⁶A methylation. Intriguingly, although Mettl3 and Mettl14 were not reliably detectable by scRNA-seq, Virma, a component of the m⁶A writer complex, was expressed at a significantly higher level in bone marrow MSCs at perinatal compared with adult stages (Figures 1E and S1F). In addition, perinatal MSCs preferentially expressed m⁶A readers, including Igf2bp1, Igf2bp2, Igf2bp3, Eif3a, Hnrnpa2b1, and Hnrnpc, with a lower level of Alkbh5, an m⁶A demethylase (but not Fto, the other m⁶A demethylase) (Figure 1E). Reverse transcription-quantitative PCR (RT-qPCR) analysis demonstrated a drastic and gradual downregulation of Mettl3 in bone marrow MSCs within two weeks after birth (Figure 1F). Consistently, perinatal MSCs expressed significantly more METTL3 protein than their adult counterparts (Figures 1G and 1H). Corresponding elevated m⁶A levels were also observed in perinatal bone marrow stromal cells (Figure 1I). Collectively, these data raised the possibility that epitranscriptomic regulation by Mettl3-mediated m⁶A modification specifically controls the generation of the bone marrow HSC niche.

Prx1-cre; Mettl3^fl/fl mice show abnormal limb bone marrow development

Prx1-cre recombines in the limb mesenchyme from E9.5 ³⁴, allowing the study of bone marrow hematopoietic niche generation starting around E17.5 ³⁵. To carefully evaluate the recombination pattern of Prx1-cre in the developing bone marrow mesenchyme, we generated Prx1-cre; Rosa26^{LSL-tdTomato/+} mice. At two weeks old, consistent with its reported recombination pattern in the limb bones ^9,11,34, Prx1-cre recombined in limb bone mesenchymal cells, including aggrecan⁺ chondrocytes, osteopontin⁺ osteolineage cells, perilipin⁺ adipocytes, and CD45/Ter119⁻PDGFRα⁺ MSCs, but not in TRAP⁺ osteoclasts or CD45⁺ hematopoietic cells (Figures S2A–S2G). We then crossed Prx1-cre mice with Mettl3^fl mice to generate Prx1-cre; Mettl3^fl/fl mice (Figure 2A). Mettl3 was efficiently deleted in limb bone marrow MSCs from two-week-old Prx1-cre; Mettl3^fl/fl mice, leading to a significant reduction in the m⁶A level on mRNA (Figures 2B and 2C). Prx1-cre; Mettl3^fl/fl mice were born at a ratio lower than expected (Figure S2H). They also displayed decreased postnatal viability compared with littermate controls, with the majority not surviving to three weeks old (Figure S2I). Prx1-cre; Mettl3^fl/fl mice had similar body size relative to controls at P0, but were stunted and much smaller later (Figure S2J), suggesting a delayed growth. Mutant mice also had shorter limbs, as well as smaller and paler limb bones (Figures S2J and S2K). Consistent with the pale appearance, hematoxylin and eosin staining revealed that the limb bones from Prx1-cre; Mettl3^fl/fl mice had few hematopoietic cells but abundant mesenchymal tissues at P0 (Figure S2L). By two weeks old, the phenotypes in the femurs from Prx1-cre; Mettl3^fl/fl mice had progressed further, with more extensive mesenchymal tissues and sparse hematopoietic cells in the central bone marrow (Figure S2M). These data suggest that Mettl3 from mesenchymal cells is required for the generation of bone marrow hematopoietic architecture during ontogeny.

Figure 2. Prx1-cre; Mettl3fl/fl mice have impaired hematopoiesis and HSC depletion in the limb bone marrow. See also Figures S2 and S3.

(A) Schematic diagram showing the breeding strategy to generate Prx1-cre; Mettl3^fl/fl (Δ/Δ) mice. Control includes Mettl3^fl/+, Mettl3^fl/fl, and Prx1-cre; Mettl3^fl/+ mice as they were phenotypically indistinguishable.

(B) Mettl3 mRNA levels in CD45/Ter119⁻PDGFRα⁺ bone marrow MSCs from two-week-old Δ/Δ and Mettl3^fl/+or Mettl3^fl/fl control mice.

(C) mRNA m⁶A modification abundance in cultured bone marrow stromal cells from two-week-old Δ/Δ and Mettl3^fl/+or Mettl3^fl/fl control mice.

(D) Bone marrow cellularity of two hindlimbs from control and Δ/Δ mice at indicated time points.

(E) Frequencies of indicated hematopoietic cell populations in the hindlimb bone marrow of two-week-old control and Δ/Δ mice.

(F and G) HSC frequencies and numbers in two hindlimb bone marrow of control and Δ/Δ mice at indicated time points.

(H) HSC frequencies in the vertebral bone marrow of two-week-old control and Δ/Δ mice.

(I and J) HSC frequencies in the livers and spleens of control and Δ/Δ mice at indicated time points.

(K) Frequencies of functional HSCs in the hindlimb bone marrow of control and Δ/Δ mice determined by limiting dilution assay.

Data are represented as mean ± SD. Each dot indicates one mouse. P values were calculated with two-tailed unpaired t-test in B-J and with chisq test in K.

The marrow in limb bones from Prx1-cre; Mettl3^fl/fl mice is defective in supporting HSCs

Consistent with the histology results (Figures S2L and S2M), limb bones from neonatal Prx1-cre; Mettl3^fl/fl mice had reduced bone marrow cellularity compared with controls (Figure 2D). There was a reduction in the frequency of B220⁺ B cells, with no difference in the frequencies of other hematopoietic cells, including CD3⁺ T cells, CD41⁺ megakaryocytic cells, Mac1⁺Gr1⁺ myeloid cells, or Ter119⁺ erythrocytes in the bone marrow from Prx1-cre; Mettl3^fl/fl mice (Figure 2E). The frequency and number of CD150⁺CD48⁻LSK HSCs were significantly reduced in the hindlimb bone marrow of Prx1-cre; Mettl3^fl/fl mice at neonatal stages, reaching a 3-fold reduction in frequency and a 16-fold reduction in number by two weeks old (Figures S3A, 2F, and 2G). Notably, in contrast to control mice where HSC number and total cellularity gradually increased in the bone marrow postnatally, HSCs and total cells were never in prominent numbers in the hindlimb bone marrow of neonatal Prx1-cre; Mettl3^fl/fl mice (Figures 2D and 2G), suggesting a defect in bone marrow HSC niche formation. In line with the limb, but not axial, bone recombination pattern of Prx1-cre ³⁴, vertebral marrow from two-week-old Prx1-cre; Mettl3^fl/fl mice did not exhibit HSC depletion but rather had a higher HSC frequency compared with controls (Figure 2H), ruling out an HSC-intrinsic defect.

HSCs expand in the fetal liver during mid-gestation and start to migrate to the spleen and then to the bone marrow perinatally ^1,5,35. P0 Prx1-cre; Mettl3^fl/fl mice had similar HSC frequency in the livers compared to controls (Figure 2I), indicating normal HSC expansion in the developing liver. Interestingly, by P5, the HSC frequency in the livers of Prx1-cre; Mettl3^fl/fl mice was significantly higher compared with littermate controls (Figure 2I). Similarly, the HSC frequency in the spleens of Prx1-cre; Mettl3^fl/fl mice was unaffected until two weeks old, when a modestly higher frequency was evident (Figure 2J). These data suggest HSC retention and redistribution at extramedullary hematopoietic sites. No significant accumulation of HSCs was observed in the peripheral blood (Figure S3B). With regard to the overall distribution of most HSCs in the body, Prx1-cre; Mettl3^fl/fl mice had a significant reduction in the portion of HSCs in the limb bones with a higher portion in other bones compared with controls (Figure S3C).

To assess the frequency of functional HSCs in the bone marrow, we performed limiting dilution assay by transplanting whole bone marrow cells from the hindlimbs of Prx1-cre; Mettl3^fl/fl or control mice, along with competitive whole bone marrow cells, into irradiated recipient mice. There was a 3.5-fold reduction in the frequency of functional HSCs within the bone marrow from hindlimbs of Prx1-cre; Mettl3^fl/fl mice compared with controls (Figures 2K, S3D, and S3E). These data suggest that loss of Mettl3 from MSCs during ontogeny leads to defects in the generation of functional bone marrow HSC niche.

Ocn-cre; Mettl3^fl/fl and Lepr-cre; Mettl3^fl/fl mice exhibit normal HSC number and function in the bone marrow

Osteolineage cells are neither a major source of HSC niche factors in the perinatal bone marrow (Figure 1B) nor a functional component of the adult bone marrow HSC niche ³⁶. Nonetheless, Prx1-cre recombines in osteoblasts ³⁴ (Figures S2B and S2F), and Prx1-cre; Mettl3^fl/fl mice appear to have osteogenesis phenotypes (Figure S2K), prompting us to formally test whether deletion of Mettl3 from osteoblasts contributes to the HSC niche defects in neonatal Prx1-cre; Mettl3^fl/fl mice. Consistent with a prior report ³⁷, we found that Ocn-cre recombined in osteopontin⁺ osteoblasts, but not other stromal cell lineages (Figures S3F–S3I). We then deleted Mettl3 from osteoblasts by generating Ocn-cre; Mettl3^fl/fl mice (Figures 3A and 3B). Two-week-old Ocn-cre; Mettl3^fl/fl mice showed normal body size and limb bone length compared with littermate controls (Figures 3C and 3D). These mice also had normal cellularity, LSK frequency and number, and HSC frequency and number in the bone marrow (Figures 3E–3I). Bone marrow cells from Ocn-cre; Mettl3^fl/fl mice had similar reconstitution capacity compared with littermate controls (Figure 3J). These data suggest that Mettl3 from osteoblasts is dispensable for HSC niche generation in the bone marrow.

Figure 3. Ocn-cre; Mettl3fl/fl and Lepr-cre; Mettl3fl/fl mice have normal hematopoiesis and normal numbers of HSCs in the bone marrow. See also Figure S3.

(A) Schematic diagram showing the breeding strategy to generate Ocn-cre; Mettl3^fl/fl mice.

(B) Mettl3 mRNA levels in tdTomato⁺ osteoblastic cells from two-week-old Ocn-cre; Rosa26^tdTomato/+ and Ocn-cre; Mettl3^fl/fl; Rosa26^tdTomato/+ mice.

(C) Representative images of two-week-old Mettl3^fl/fl and Ocn-cre; Mettl3^fl/fl mice.

(D) Representative images of femurs and tibias from two-week-old Mettl3^fl/fl and Ocn-cre; Mettl3^fl/fl mice.

(E-I) Bone marrow cellularity (E), LSK frequency (F), HSC frequency (G), LSK number (H), and HSC number (I) in two hindlimbs from two-week-old Mettl3^fl/fl and Ocn-cre; Mettl3^fl/fl mice.

(J) Donor-derived blood cells in competitive transplantation assays. Data were pooled from three independent experiments with three donor pairs.

(K) Schematic diagram showing the breeding strategy to generate adult Lepr-cre; Mettl3^fl/fl mice.

(L) Mettl3 mRNA levels in the CD45/Ter119⁻PDGFRα⁺ bone marrow MSCs from adult Lepr-cre; Mettl3^fl/fl and littermate Mettl3^fl/fl control mice.

(M) Representative images of adult Mettl3^fl/fl and Lepr-cre; Mettl3^fl/fl mice.

(N) Representative images of femurs and tibias from adult Mettl3^fl/fl and Lepr-cre; Mettl3^fl/fl mice.

(O-R) Bone marrow cellularity (O), CD45/Ter119⁻PDGFRα⁺ bone marrow MSC frequency (P), LSK frequency (Q), and HSC frequency (R) in two hindlimbs of adult Mettl3^fl/fl and Lepr-cre; Mettl3^fl/fl mice.

(S) Donor-derived blood cells in competitive transplantation assays. Data were pooled from two independent experiments with two donor pairs.

Data are represented as mean ± SD. P values were calculated with two-tailed unpaired t-test, except in J and S, where two-way ANOVA followed by Bonferroni’s multiple comparisons were used.

Mettl3 expression in bone marrow MSCs is drastically downregulated postnatally (Figure 1F), raising the possibility that Mettl3 is not required for the maintenance of the niche. As Lepr-cre recombines efficiently in bone marrow MSCs at adult but not early postnatal stages (in 10% of LepR⁺ cells at P4 and 30% of LepR⁺ cells at P14) ^8,38, we generated Lepr-cre; Mettl3^fl/fl mice to test whether Mettl3 is required for adult HSC niche maintenance (Figure 3K). Mettl3 was efficiently deleted from the bone marrow MSCs of adult Lepr-cre; Mettl3^fl/fl mice (Figure 3L). Adult Lepr-cre; Mettl3^fl/fl mice had normal body size and limb bone length compared with littermate controls (Figures 3M and 3N). The total cellularity as well as the frequencies of CD45/Ter119⁻PDGFRα⁺ MSCs, LSKs, and HSCs in the bone marrow from Lepr-cre; Mettl3^fl/fl mice were also normal (Figures 3O–3R). Bone marrow cells from Mettl3^fl/fl control and Lepr-cre; Mettl3^fl/fl mutant mice gave comparable levels of donor-derived peripheral blood cells in competitive transplantation (Figure 3S). We further generated Lepr-cre; Mettl3^fl/− mice, aiming to delete Mettl3 more efficiently as recombination of only one floxed allele is required to achieve Mettl3 deletion in this model (Figure S3J). Lepr-cre; Mettl3^fl/− mice had normal hematopoiesis and HSC numbers in the bone marrow compared with littermate Mettl3^fl/− controls (Figures S3K–S3N). These data suggest that Mettl3 in MSCs is not required for HSC niche maintenance in the bone marrow.

Bone marrow stromal composition is disturbed in Prx1-cre; Mettl3^fl/fl mice

To systematically understand how the bone marrow HSC niche generation is impaired in Prx1-cre; Mettl3^fl/fl mice, we performed scRNA-seq on CD45/Ter119⁻ bone marrow stromal cells from the limb bones of these mice along with controls at two weeks old (Figures 4A and S4A), when HSCs are most severely depleted in the mutants (Figure 2). After removing hematopoietic and cycling cells, we identified 8 clusters from the aggregated stromal cells (Figures 4B, 4C, and S4B). Two clusters were identified as MSCs (cluster 1, MSC_Adipo and cluster 2, MSC_Osteo). Consistent with previous scRNA-seq data ^38,39, cells of the MSC_Adipo cluster expressed high levels of Adipoq and Lpl, while cells of the MSC_Osteo cluster expressed high levels of Clec11a (osteolectin) and Sp7 (Figure 4D). Two osteolineage clusters (cluster 3, osteoblasts expressing Ocn (Bglap) and Bglap2, and cluster 4, osteocytes expressing Dmp1), and one chondrocyte cluster (cluster 5, expressing Acan (aggrecan) and Col2a1) were also identified (Figures 4B–4D and S4C). The endothelial cell (EC) cluster (expressing Cdh5 (VE-cadherin) and Kdr) was further divided into sinusoidal (cluster 6, EC_sinusoidal, Stab2^high and Flt4^high) and arteriolar (cluster 7, EC_Arteriolar, Ly6a^high and Cd34^high) clusters (Figures 4B–4D and S4C). In addition, one cluster (cluster 8, fibroblast cluster) expressed fibroblast markers Col3a1 and Fgfr4 (Figures 4B–4D and S4C). Although the frequency of total endothelial cells (EC_Sinusoidal and EC_Arteriolar) was largely comparable, a higher frequency of the EC_Arteriolar cluster was in the bone marrow from Prx1-cre; Mettl3^fl/fl mice compared with controls, accompanied by a lower frequency of the EC_Sinusoidal cluster (Figure 4E). Bone marrow from Prx1-cre; Mettl3^fl/fl mice contained higher frequencies of cells in the MSC_Osteo and osteolineage (osteoblast and osteocyte) clusters, but a lower frequency of cells in the MSC_Adipo cluster (Figures 4C and 4E). These data suggest that the cellular composition of the bone marrow stroma is distorted when Mettl3 is deleted from the developing MSCs.

Figure 4. Prx1-cre; Mettl3fl/fl mice have disturbed stromal cell composition in the limb bone marrow. See also Figure S4.

(A) Schematic diagram showing scRNA-seq analysis on sorted bone marrow stromal cells from control and Prx1-cre; Mettl3^fl/fl (Δ/Δ) mice. Cells from six control and seven Δ/Δ mice were sequenced in two independent scRNA-seq experiments.

(B and C) t-distributed Stochastic Neighbor Embedding (t-SNE) visualization showing cell clusters of bone marrow stroma identified by scRNA-seq. Data from control and Δ/Δ mice were aggregated in B and separated in C. A total of 5,726 cells, including 1,870 cells from control and 3,856 cells from Δ/Δ mice, are shown.

(D) Dot plot showing the expression patterns of indicated genes in each cluster.

(E) Quantification of stromal cell composition changes between control and Δ/Δ mice. Error bars represent 95% confidence intervals of binomial fit mean using genotype (control or Δ/Δ) as the predictive factor. **, p<0.01.

(F) t-SNE plots showing the expression patterns of Cxcl12 and Scf in MSCs. The areas enclosed by dotted lines indicate MSCs (including the MSC_Adipo and MSC_Osteo clusters in B).

(G) Violin plots showing the expression levels of indicated genes in MSCs from control and Δ/Δ mice. Hematopoietic niche-associated genes, adipo-, osteo-, and chondro-markers are shadowed in red, blue, purple, and yellow, respectively.

Similar to the adult bone marrow ³⁹, Cxcl12 and Scf, two critical HSC niche factors, were mainly expressed by MSCs (clusters 1 and 2) in the two-week-old bone marrow (Figures 4F and S4D). MSCs from Prx1-cre; Mettl3^fl/fl mice expressed lower levels of Cxcl12, Scf, other MSC markers (e.g., Lepr and Pdgfra), and adipogenesis markers (e.g., Adipoq, Lpl, Cebpa, and Pparg) compared with controls (Figures 4G and S4D). Furthermore, they expressed higher levels of osteogenesis markers (e.g., Col1a1, Ocn, Bglap2, Spp1 (osteopontin), Sp7, and Alpl) but not chondrolineage markers (e.g., Acan) compared with controls (Figures 4G and S4D). Among all stromal cells identified by scRNA-seq, MSCs from Prx1-cre; Mettl3^fl/fl mice showed the most differentially expressed genes (DEGs) (159 genes, mainly MSC and osteogenesis markers) compared with controls (Table S1). Only 13 DEGs were identified between sinusoidal endothelial cells with no differentially expressed genes between arteriolar endothelial cells. Some arteriolar genes, such as Cd34 and Fbln2, were expressed at higher levels in sinusoidal endothelial cells from Prx1-cre; Mettl3^fl/fl mice compared with controls (Table S1), suggesting a possible endothelial fate transition. We only identified 5 DEGs in osteoblasts (Hba-a1, Cxcl12, Igfbp5, Col11a2 downregulation and Tnc upregulation) and 1 DEG in chondrocytes (Zfp36l1 upregulation) from Prx1-cre; Mettl3^fl/fl mice (Table S1). Overall, these data suggest that MSCs losing their niche cell identity and skewing towards an osteolineage fate are the major alterations in the bone marrow niche of Prx1-cre; Mettl3^fl/fl mice.

HSC niche generation is compromised in the bone marrow of Prx1-cre; Mettl3^fl/fl mice

Our scRNA-seq analysis suggests that MSCs have biased differentiation to osteolineage cells in Prx1-cre; Mettl3^fl/fl mice (Figure 4G). Consistently, immunostaining revealed that these mice had significantly more osteopontin⁺ osteolineage cells in the bone marrow (Figures S5A and S5B), with normal numbers of chondrocytes and endothelial cells (Figures S5C–S5F). This was accompanied by more osteoclasts and fewer hematopoietic cells (Figures S5G–S5J). In addition, fewer lipid droplet-containing adipocytes were in the bone marrow of Prx1-cre; Mettl3^fl/fl mice (Figures S5K–S5N), although adipocytes were rare in the control bone marrow as well.

We used Cxcl12-DsRed reporter mice ⁹ to assess the distribution of MSCs in situ. Significantly fewer Cxcl12-DsRed⁺ cells were in the bone marrow of P0 Prx1-cre; Mettl3^fl/fl; Cxcl12^DsRed/+ mice compared with controls (Figures S5O and S5P). By two weeks old, Cxcl12-DsRed⁺ MSCs were abundantly distributed throughout the control bone marrow (Figure 5A). However, these cells were restricted to only a few small regions in the bone marrow from the mutant mice (Figures 5A and 5B). Lineage-tracing using a Rosa26^LSL-ZsGreen reporter revealed that Prx1-cre lineage cells were distributed throughout the bone marrow of Prx1-cre; Mettl3^fl/fl; Cxcl12^DsRed/+; Rosa26^{LSL-ZsGreen/+} mice (Figures 5A and 5B), suggesting that the mesenchymal lineage cells are not depleted as a result of Mettl3 loss. While most ZsGreen⁺ mesenchymal lineage cells were positive for Cxcl12-DsRed in the control bone marrow, the majority of the same cell population from Prx1-cre; Mettl3^fl/fl; Cxcl12^DsRed/+; Rosa26^{LSL-ZsGreen/+} mice were negative for Cxcl12-DsRed (Figures 5A and 5B). Flow cytometry also revealed a significant reduction in the frequency of CD45/Ter119⁻Cxcl12-DsRed⁺ cells in the bone marrow of Prx1-cre; Mettl3^fl/fl; Cxcl12^DsRed/+ mice compared with controls (Figures 5C and 5D).

Figure 5. Prx1-cre; Mettl3fl/fl mice have compromised hematopoietic niche formation in the bone marrow. See also Figure S5.

(A) Representative confocal images of femur sections from two-week-old Cxcl12^DsRed/+; Rosa26^{LSL-ZsGreen/+} and Prx1-cre; Mettl3^fl/fl; Cxcl12^DsRed/+; Rosa26^{LSL-ZsGreen/+} mice.

(B) Quantification of Cxcl12-DsRed⁺ area in the bone marrow (top) and in ZsGreen⁺ area (bottom).

(C and D) Representative flow cytometry plots (C) and quantification (D) of Cxcl12-DsRed⁺ stromal cell frequencies in the hindlimb bone marrow of two-week-old Cxcl12^DsRed/+ and Prx1-cre; Mettl3^fl/fl; Cxcl12^DsRed/+ mice.

(E and F) Representative flow cytometry plots (E) and quantification (F) of bone marrow CD45/Ter119⁻PDGFRα⁺ cell frequencies in hindlimbs of two-week-old control and Δ/Δ mice.

(G) RT-qPCR analysis of expression levels of indicated genes in bone marrow CD45/Ter119⁻ PDGFRα⁺ cells from two-week-old control and Δ/Δ mice.

(H and I) Representative flow cytometry plots (H) and quantification (I) of CXCL12 expression in hindlimb bone marrow CD45/Ter119⁻PDGFRα⁺ cells from two-week-old control and Δ/Δ mice.

Data are represented as mean ± SD. P values were calculated with two-tailed unpaired t-test. Δ/Δ, Prx1-cre; Mettl3^fl/fl mice.

Using a set of independent markers for bone marrow MSCs ¹⁴, we also observed a lower frequency of CD45/Ter119⁻PDGFRα⁺ cells in the bone marrow of Prx1-cre; Mettl3^fl/fl mice compared with controls (Figures 5E and 5F). These bone marrow CD45/Ter119⁻PDGFRα⁺ cells from Prx1-cre; Mettl3^fl/fl mice expressed significantly lower levels of Lepr, Pdgfra, Cxcl12, and Scf (Figure 5G). The level of CXCL12 protein in CD45/Ter119⁻PDGFRα⁺ cells from Prx1-cre; Mettl3^fl/fl mice was also significantly reduced (Figures 5H and 5I). These data suggest that bone marrow HSC niche generation is compromised in Prx1-cre; Mettl3^fl/fl mice.

Mettl3 regulates bone marrow HSC niche formation by repressing Klf2

Mettl3 expression in MSCs is gradually downregulated from P2 (Figures 1F–1I), suggesting that METTL3 may deposit much of m⁶A onto target mRNAs at the early postnatal stage. To unravel the molecular mechanisms by which Mettl3 regulates bone marrow MSC fate and HSC niche generation, we sorted CD45/Ter119/CD31⁻Cxcl12-DsRed⁺ MSCs from the limb bone marrow of P2 Cxcl12^DsRed/+ mice and performed m⁶A-tagged mRNA immunoprecipitation sequencing (meRIP-seq) for rare cells (Figure 6A), as previously described ⁴⁰. Compared with input, 1,928 genes were significantly enriched in the anti-m⁶A antibody-bound fraction (Figures 6B, S6A, and S6B, and Table S2). Consistent with the notion that RNA regulatory mechanisms are enriched in perinatal MSCs (Figures S1B–S1D), GO analysis of the putative m⁶A targets identified enrichment of pathways involved in RNA splicing and mRNA processing (Figure S6C and Table S3). Considering that m⁶A modification may control the abundance of target mRNAs by regulating their decay, we intersected the meRIP-seq and scRNA-seq datasets (Figure 6C). Ten genes were both m⁶A targets and differentially expressed between control and Mettl3-deficient MSCs. We further focused on transcription factors as they are key cell fate determinants. Klf2 was the only transcription factor that was an m⁶A target and also differentially expressed between control and Mettl3-deficient MSCs (Figure 6C). Previous studies demonstrated that Klf2 overexpression promotes osteolineage differentiation but represses adipogenesis in vitro ^41–44, similar to the phenotypes in Prx1-cre; Mettl3^fl/fl mice, leading us to carefully examine Klf2. MeRIP-qPCR also revealed that Klf2 transcript was significantly enriched in anti-m⁶A antibody-bound fraction compared with input or m⁶A antibody-unbound fractions (Figure 6D). These data suggest that Klf2 is a direct target of m⁶A in MSCs. scRNA-seq and RT-qPCR data showed that Klf2 was expressed higher in Mettl3-deficient MSCs compared with controls (Figures 6E and 6F). Using KLF2-GFP fusion protein reporter mice ⁴⁵, we found that the level of KLF2 protein was significantly higher in Mettl3-deficient MSCs compared with controls (Figures 6G and 6H). Klf2 mRNA was more stable in cultured Mettl3-deficient bone marrow stromal cells (Figures S6D and S6E). Using a dCas13b system ^46–48, we targeted m⁶A demethylase FTO or ALKBH5 to Klf2 mRNA in cultured bone marrow stromal cells (Figure S6F). This led to a significantly reduced m⁶A level on Klf2 mRNA and upregulation of Klf2 mRNA expression (Figure S6G and S6H). Taken together, these data suggest that Mettl3-mediated m⁶A modification inhibits Klf2 expression by promoting mRNA degradation.

Figure 6. Mettl3 represses Klf2 expression in bone marrow MSCs. See also Figure S6.

(A) Schematic diagram showing meRIP-seq on CD45/Ter119/CD31⁻DsRed⁺ cells sorted from hindlimbs of P2 Cxcl12^DsRed/+ mice.

(B) Heatmap showing mRNA abundance. A total of 9,114 genes with mean abundance in at least one fraction above the median abundance of all detectable genes are shown to reduce uncertainty.

(C) Venn diagram showing the number of transcription factor-encoding genes that were enriched (log₂FoldChange > 2 and adjusted p value < 0.05) in anti-m⁶A-bound compared with input fractions and/or that were differentially expressed (|log₂FoldChange| > 1 and adjusted p value < 0.05) between control and Δ/Δ MSCs revealed by scRNA-seq.

(D) MeRIP-qPCR analysis validating that Klf2 mRNA is enriched in the anti-m⁶A-bound fraction. Klf2 mRNA levels in indicated fractions were normalized to that of Gapdh, which is not enriched in the anti-m⁶A-bound fraction revealed by our meRIP-seq data.

(E) Klf2 mRNA levels in bone marrow MSCs (including the MSC_Adipo and MSC_Osteo clusters) from control and Prx1-cre; Mettl3^fl/fl (Δ/Δ) mice revealed by scRNA-seq.

(F) Klf2 mRNA levels in bone marrow CD45/Ter119⁻PDGFRα⁺ cells from control and Δ/Δ mice revealed by RT-qPCR.

(G and H) Expression of KLF2 protein in bone marrow CD45/Ter119⁻PDGFRα⁺ cells from control and Δ/Δ mice revealed by a KLF2-GFP reporter. Representative flow cytometry plots and quantification of KLF2-GFP MFI are shown in (G) and (H), respectively.

(I) Schematic diagram showing experimental procedures to overexpress Klf2 or GFP control in cultured bone marrow stromal cells from two-week-old mice.

(J) The expression levels of indicated genes in cultured bone marrow stromal cells transduced with GFP control or Klf2-overexpressing vector.

(K and L) Representative flow cytometry plots (K) and quantification (L) of Cxcl12-DsRed expression in cultured bone marrow stromal cells from Cxcl12^DsRed/+ mice transduced with GFP- or Klf2-overexpressing vector.

Data are presented as mean ± SD. P values were calculated with two-tailed unpaired t-test, except in D, where one-way ANOVA followed by Dunnett’s multiple comparisons test was applied.

Cultured Mettl3-deficient bone marrow stromal cells expressed higher levels of Klf2 and osteolineage genes (Col1a1, Runx2, Ocn, Bglap2, and Alpl), but lower levels of hematopoietic niche factors (Cxcl12 and Scf) and adipolineage genes (Adipoq and Pparg) compared with controls (Figures S6I). Overexpression of Klf2 in bone marrow stromal cells led to increased levels of osteolineage genes and decreased levels of hematopoietic niche factor and adipolineage genes (Figures 6I and 6J). Using bone marrow stromal cells cultured from the Cxcl12^DsRed/+ reporter mice, we found that Klf2 overexpression led to a lower level of Cxcl12-DsRed (Figures 6I, 6K, and 6L). These data collectively suggest that loss of Mettl3 results in upregulation of Klf2, which in turn leads to increased expression of osteolineage genes and decreased expression of HSC niche factors in bone marrow stromal cells.

Deletion of Klf2 rescues the HSC niche defects in the bone marrow of Prx1-cre; Mettl3^fl/fl mice

To test whether Klf2 is a functional target of m⁶A in MSCs during bone marrow niche formation in vivo, we generated Prx1-cre; Mettl3^fl/fl; Klf2^fl/fl mice and littermate controls, including no deletion controls, Mettl3 single deletion (Mettl3-del), and Klf2 single deletion (Klf2-del) (Figures 7A and S7A). Compared with Mettl3-del, Prx1-cre; Mettl3^fl/fl; Klf2^fl/fl mice were born at higher numbers than expected (Figure S7B). The shortened limbs, stunted growth, and reduced survival rate of Mettl3-del mice were not rescued in Prx1-cre; Mettl3^fl/fl; Klf2^fl/fl mice (Figure 7B). However, the hindlimb bones from Prx1-cre; Mettl3^fl/fl; Klf2^fl/fl mice were redder in color compared with those from Mettl3-del mice (Figure 7B), suggesting that the bone marrow niche defects and hematopoiesis may be rescued by Klf2 deletion.

Figure 7. Deletion of Klf2 rescues bone marrow HSC niche defects in Prx1-cre; Mettl3fl/fl mice. See also Figure S7.

(A) Schematic diagram showing the breeding strategy to generate no deletion Control, Klf2 deletion (Klf2-del), Mettl3 deletion (Mettl3-del), and Mettl3 and Klf2 double deletion (Double-del) mice. See Figure S7A for detailed information on the genotypes.

(B) Representative images of femurs and tibias from two-week-old Control, Klf2-del, Mettl3-del, and Double-del mice.

(C and D) Frequencies of and expression levels of indicated genes in CD45/Ter119⁻PDGFRα⁺ cells in the hindlimb marrow of two-week-old Control, Klf2-del, Mettl3-del, and Double-del mice.

(E-H) Cellularity (E), B220⁺ cell frequency (F), HSC frequency (G), and HSC number (H) in the marrow of two hindlimbs from two-week-old Control, Klf2-del, Mettl3-del, and Double-del mice.

(I) Donor-derived blood cell populations in recipients transplanted with 500,000 whole bone marrow cells from the hindlimb of control and Klf2-del mice along with 500,000 competitor cells. Data were pooled from two independent experiments with two donor pairs.

(J) Donor-derived blood cell populations in recipients transplanted with 500,000 whole bone marrow cells from the hindlimb of Mettl3-del and Double-del mice along with 500,000 competitor cells. Data were pooled from two independent experiments with two donor pairs. Data are presented as mean ± SD. P values were calculated with two-tailed unpaired t-test, except in I and J, where two-way ANOVA followed by Bonferroni’s multiple comparisons test was applied.

Immunostaining revealed that the bone marrow hematopoietic architecture was notably rescued in Prx1-cre; Mettl3^fl/fl; Klf2^fl/fl mice compared with Mettl3-del mice, with less osteogenesis and more hematopoiesis (Figures S7C–S7H). Consistently, two-week-old Prx1-cre; Mettl3^fl/fl; Klf2^fl/fl mice had a significantly higher frequency of bone marrow CD45/Ter119⁻ PDGFRα⁺ MSCs compared with Mettl3-del mice (Figures 7C and S7I). Bone marrow MSCs from Prx1-cre; Mettl3^fl/fl; Klf2^fl/fl mice expressed significantly higher levels of Cxcl12, Scf, and Adipoq, but lower levels of Col1a1, compared with those from Mettl3-del mice (Figure 7D). These data suggest that both the number and quality of the niche cells in Mettl3-del mice are significantly rescued by Klf2 deletion. Klf2-del mice had normal bone marrow cellularity and frequencies of differentiated hematopoietic cells compared with no deletion control mice (Figures 7E and S7J). However, deletion of Klf2 and Mettl3 in Prx1-cre; Mettl3^fl/fl; Klf2^fl/fl mice led to higher cellularity and B cell frequency in the bone marrow compared with Mettl3 deletion alone (Figures 7E and 7F). Although the bone marrow HSC frequency in Klf2-del mice was normal, the bone marrow HSC frequency in Prx1-cre; Mettl3^fl/fl; Klf2^fl/fl mice was significantly higher compared with Mettl3-del mice, reaching a level similar to no deletion controls (Figures 7G and S7K). This led to a significantly higher HSC number in the bone marrow of Prx1-cre; Mettl3^fl/fl; Klf2^fl/fl mice compared with Mettl3-del mice (Figure 7H). While the bone marrow cells from Klf2-del mice reconstituted normally (Figure 7I), the bone marrow cells from Prx1-cre; Mettl3^fl/fl; Klf2^fl/fl mice had a significantly higher reconstitution capacity than those from Mettl3-del mice (Figure 7J). These results suggest that Klf2 is a functional target of Mettl3 in regulating bone marrow MSC fate and HSC niche generation.

Discussion

Previous studies have identified factors that are preferentially expressed in MSCs and required for adult bone marrow HSC niche maintenance ^27–31. In contrast, the molecular mechanisms regulating the generation of the niche remained largely unexplored. A prior study suggests that Foxc1 may play an important role in HSC niche generation as its conditional deletion from MSCs leads to HSC niche defects in three-week-old mice ²⁷. Interestingly, similar HSC niche defects were also observed when Foxc1 was conditionally deleted from adult mice ²⁷, suggesting that Foxc1 is required for niche maintenance. This raises the question of whether the niche defects in the three-week-old mutants are due to the role of Foxc1 in niche maintenance or generation. Furthermore, Foxc1 is minimally expressed by perinatal bone marrow MSCs compared with adults (Figure 1C), suggesting a limited role in niche generation. Instead, we searched for genes that are preferentially expressed during niche generation at the perinatal stage. Genes mediating m⁶A-mediated epitranscriptomic regulation are preferentially expressed in perinatal MSCs (Figure 1). Mettl3 expression in MSCs and the resulting m⁶A level are downregulated in adult versus perinatal stages (Figures 1F–1I), suggesting their specific role for HSC niche generation but not maintenance. Consistent with the expression data, our functional study demonstrates that the Mettl3-dependent epitranscriptomic mechanism is indeed specifically required for the generation but not maintenance of the bone marrow HSC niche.

Wu et al. showed that adult Prx1-cre; Mettl3^fl/fl mice displayed impaired osteodifferentiation ⁴⁹, an osteogenesis phenotype distinct from what we observed here during development. Besides the different stages of analysis, one technical difference is that their Mettl3^fl allele was generated by inserting loxp sites flanking exons 2 and 3 using CRISPR/Cas9, while our Mettl3^fl allele was generated by inserting loxp sites flanking exon 4 using conventional gene targeting in embryonic stem cells. Exon 4 encodes the RNA recognition-zinc finger domain required for Mettl3 methyltransferase activity, and cells with exon 4 deletion have virtually no m⁶A on mRNAs ^50,51. Similar strategies targeting exon 4 have been used to generate other independent Mettl3 conditional null alleles ^52,53. In contrast, targeting exon 2 via CRISPR/Cas9 still retains ~40% of the m⁶A signal due to splicing that bypasses the mutation and produces functional METTL3 protein, suggesting that targeting exon 2 may not lead to a null of Mettl3 ^50,54. Thus, it is possible that the distinct phenotypes in osteogenesis may come from the differences in the efficiency of generating a null allele of Mettl3.

Our findings fill a critical knowledge gap by unambiguously discovering a pathway that specifically regulates the generation of the HSC niche and demonstrating that distinct molecular mechanisms regulate the generation and maintenance of a stem cell niche. It is likely that these molecular distinctions reflect different functional requirements of the niche cells at different stages (e.g., fate decision during ontogeny versus cytokine secretion in adults). Our study thus provides a conceptual framework to understand the generation of the HSC niche.

Limitations of the Study

Deletion of Klf2 did not completely rescue the marrow and bone phenotypes from Mettl3 deficiency (Figure 7), suggesting that HSC niche generation and limb development are also mediated by m⁶A targets beyond Klf2. Further studies are required to uncover other functional m⁶A targets in MSCs. Molecular profiling of rare primary cells is challenging but critical for understanding the mechanisms that regulate HSCs in vivo. The scRNA-seq (Figures 1 and 4) and meRIP-seq (Figure 6) were performed using available methods for rare cells, with inherited technical limitations. While more advanced methods are needed to more robustly profile rare cells, independent approaches to validate the conclusions from these profiling analyses are required. Our scRNA-seq revealed composition changes in the endothelial cell compartment from the bone marrow of Prx1-cre; Mettl3^fl/fl mice (Figure 4E). This may reflect the critical role of MSCs in regulating endothelial cells. Further work is needed to understand how MSCs regulate the fate and function of other niche components in the bone marrow.

STAR Methods

RESOURCE AVAILABILITY

Lead contact

Further information and reagent requests should be directed to and will be fulfilled by the lead contact, Lei Ding (ld2567@cumc.columbia.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• Single-cell RNA-seq data in Figures 1 and S1 are from GSE178951. Single-cell RNA-seq and meRIP-seq data generated in this study have been deposited at GEO under accession number GSE217057.

• No original code was generated in this study.

• Any additional information required to reanalyze the data reported in this paper is available from the Lead Contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mice

Prx1-cre 34, Lepr-cre 56, Mettl3^{fl 51}, Cxcl12^{DsRed 9}, Klf2^{gfp 45}, and Klf2^{fl 45} mice were described previously. Ocn-cre 37, B6.SJL-Ptprc^aPepc^b/BoyJ, Rosa26^LSL-ZsGreen, and Rosa26^{LSL-tdTomato 57} mice were obtained from the Jackson Laboratory. Mice of both genders were used. All mice were maintained in C57BL/6 background and housed in specific pathogen-free conditions with ad libitum access to food and water at Columbia University Irving Medical Center. Randomization or blinding was not used. Mouse experiments were in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by the Institutional Animal Care and Use Committee (IACUC) of Columbia University.

Cell culture

Bone marrow stromal cells were isolated as previously described ⁵⁸. Briefly, after the collected femurs and tibias were cut at both ends, they were placed into an end-removed 200-μl pipette tip. The bone-containing tip was put into a 1.5-ml centrifuge tube pre-filled with 100 μl stromal cell culture medium (αMEM with 10% FBS, 100 U/ml penicillin, and 100 ug/ml streptomycin) and then centrifuged at 10,000 g for 15 seconds at room temperature. Single-cell suspension was prepared by pulling the cell pellets up and down slowly with a 25-gauge needle and then filtered with a 70-μm filter. The cell suspension was seeded at a density of 1×10⁶ cells/cm² in stromal cell culture media. 24 hours later, they were washed with PBS twice and fresh medium was added. Thereafter, the medium was changed every two days until the cells were split when confluent. CD45/Ter119⁺ hematopoietic cells were removed by FACS. To compare Klf2 mRNA decay rates, cultured bone marrow stromal cells from control or Prx1-cre; Mettl3^fl/fl mice were treated with 5 μM actinomycin D (Sigma Cat#A9415) and collected for RT-qPCR analysis at indicated time points. HEK293T cells for virus production were from ATCC (Cat# CRL-3216) and cultured in DMEM with 2 mM L-glutamine, 1 mM sodium pyruvate, and 10% FBS.

METHOD DETAILS

Flow cytometry

For staining of hematopoietic cells, bone marrow cells were isolated by flushing the limb bones or by crushing the limb bones or vertebrae with mortar and pestle in Ca²⁺- and Mg²⁺-free HBSS with 2% BS. To isolate liver and spleen cells, the tissues were crushed between two glass slides. Single-cell suspension was generated by passing through a 25-gauge syringe several times and filtering with a 70-μm filter. Lineage markers (CD2 (RM2-5), CD3 (17A2), CD5 (53–7.3), CD8 (53–6.7), Ter119 (TER-119), B220 (6B2), and Gr1 (8C5)), Sca-1 (E13-161.7), c-Kit (2B8), CD150 (TC15-12F12.2), and CD48 (HM48-1) (all from BioLegend) were used for HSC staining. B220, CD3, CD41 (MWReg30), Gr1, Mac1 (M1/70), and Ter119 were used for mature lineage cell staining. For staining of bone marrow stromal cells, bone marrow was flushed into Ca²⁺- and Mg²⁺-free HBSS with 2% BS. Then the marrow was digested with Collagenase IV (200 U/ml) and DNase I (200 U/ml) for 30 minutes in a 37 °C water bath. After being filtered with a 70-μm filter, cells were stained with anti-CD45 (30-F11), Ter119, and anti-PDGFRα (APA5) antibodies. Calcein-AM (BioLegend, Cat#425201) was applied to exclude dead cells during sorting for single-cell RNA sequencing. Osteolineage cells were dissociated as preciously described ⁵⁹. Briefly, hindlimb bones from two-week-old Ocn-cre; Rosa26^tdTomato/+ or Ocn-cre; Mettl3^fl/fl; Rosa26^tdTomato/+ mice were minced with scissors and then digested with Liberase (Roche, Cat#5401119001) for 30 minutes at 37 °C on an Eppendorf Thermomixer. After the cells were filtered through a 70-μm filter, the remaining tissues were digested again with Liberase as described above. CD45/Ter119/CD31⁻ tdTomato⁺ cells were sorted for RT-qPCR analysis. For intracellular staining, cells were fixed and permeabilized using the eBioscience Transcription Factor Staining kit (Thermo Fisher Scientific, Cat#00-5523-00) after cell surface antibody staining following the manufacturer’s instructions. Cells were then stained with anti-CXCL12 (R&D Systems, Cat#IC350F) or mouse IgG1k (BioLegend, MOPC-21, Cat#400101), and anti-METTL3 (BosterBio, Cat#A01758-1) or Rabbit IgG (CST, Cat#3900S) antibody in eBioscience permeabilization buffer and analyzed by flow cytometry. Flow cytometry was performed on BD Celesta or BD SORP FACSAria and analyzed with BD FACSDiva or FlowJo (v10.4). At least three biological replicates were analyzed without randomization or blinding. No data were excluded.

Immunofluorescent staining

For immunostaining on bone sections, limb bones were fixed in 4% PFA at 4 °C overnight. Bones were then decalcified with 10% EDTA for 72 hours and then dehydrated with 30% sucrose overnight. Samples were snap-frozen in OCT solution (Fisher Scientific, Cat#23-730-571) with liquid nitrogen and stored at −80 °C. Bones were sectioned by cryostat using the CryoJane tape-transfer system (Instrumedics). Bone sections were dried overnight at room temperature and stored at −80 °C. Sections were rehydrated in PBS for 5 minutes before immunostaining. The sections were first blocked with blocking buffer (1% goat serum in PBS) for 30 minutes before the primary antibody was added. Primary antibodies were incubated overnight at 4 °C. On the next day, after adequate wash with PBS, the sections were incubated with secondary antibody (with DAPI) for 1 hour at room temperature. For LipidTox and Nile Red staining, staining was performed on whole-mount bone, instead of bone sections. Briefly, bones were fixed with 4% PFA overnight and then snap-frozen in OCT solution with liquid nitrogen. Bones were trimmed longitudinally on a cryostat until about half of the bone was trimmed. Then the remaining bones were washed with PBS to remove the OCT. Afterward, trimmed bones were fixed with 4% PFA for 30 minutes at 4 °C. Nile Red (Sigma, Cat#N3013) or LipidTox (Invitrogen, Cat#H34476) and DAPI were applied to the bones and stained for 10 minutes in dark at room temperature. Slides were mounted with anti-fade mounting medium (Vector Laboratories, Cat#H-1400) and images were acquired with a Zeiss Axioscan slide scanner, Zeiss LSM780 confocal microscope, or Leica SP8 confocal microscope. Images were processed with Leica SP8 and Fiji (v2.3.0). For quantification of immunostaining images, the positively stained areas within the bone marrow (the areas within the dotted white lines in the representative images) were evaluated with Fiji. The data were normalized to control mice. The following primary antibodies were used: aggrecan (Millipore, Cat#AB1031), osteopontin (R&D, Cat#AF808), perilipin (CST, Cat#9349S), VE-cadherin (R&D, Cat#BAF1002), CD45 (BioLegend, Cat#103106), and TRAP (Abcam, Cat#ab185716). At least three biological replicates were analyzed without randomization or blinding. No data were excluded.

MeRIP-qPCR and meRIP-sequencing

MeRIP-qPCR and meRIP-sequencing were performed as previously described ⁴⁰. Briefly, about 4,000 CD45/Ter119/CD31⁻Cxcl12-DsRed⁺ cells were sorted from P2 Cxcl12^DsRed/+ bone marrow. mRNA was isolated from sorted cells using Oligo-(dT)25 magnetic Dynabeads (ThermoFisher Scientific, Cat#61011) according to the manufacturer’s instructions. 6% of eluted RNA was saved as the input fraction. 1.25 μg of anti-m⁶A antibody (Synaptic Systems, Cat#202003) or IgG was pre-bound to Protein A/G magnetic beads (Millipore, Cat#16-663X) in IP buffer (20-mM Tris pH 7.5, 140-mM NaCl, 1% NP-40, 2-mM EDTA) for 1 hour at 4 °C.

Sample RNA was incubated with antibody-bound Protein A/G beads for 2 hours at 4 °C. The beads were then washed twice in low-salt-wash buffer (10 mM Tris, pH 7.5, 5 mM EDTA), twice with high-salt-wash buffer (20 mM Tris, pH 7.5, 1 M NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, 1 mM EDTA), and twice with RIPA buffer (20 mM Tris, pH 7.5, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, 1 mM EDTA). All wash solutions were collected as the “unbound” fraction. RNA was eluted from the beads by incubating with 50 μl 20 mM N6-methyladenosine 5-monophosphate sodium salt (Sigma-Aldrich, Cat#M2780) for 1 hour at 4 °C and washed with IP buffer three times. The elution and wash solutions were collected as the “bound” fraction. Following ethanol precipitation, input, unbound, and bound fractions were reverse transcribed, amplified, tagmented, and indexed following Smart-seq2 method ⁶⁰. Sequencing was performed on NextSeq 500 with paired-end 150-bp read length at Columbia Genome Center following the manufacturer’s instructions. For meRIP-seq data analysis, raw reads were mapped to the mouse genome (mm10) using RSubread (v2.10.4) ⁶¹. Due to low RNA input for library preparation, potential PCR duplicates were removed with Picard (http://broadinstitute.github.io/picard) as previously described ⁴⁰. Raw counts were generated with the featureCounts function of RSubread. RPKM calculation and differential abundance analysis were performed with edgeR (3.38.1) ⁶². For meRIP-qPCR, enrichment of m⁶A-containing mRNAs was determined by quantitative PCR relative to Gapdh. Biological replicates were analyzed without randomization or blinding. No data were excluded.

Single-cell RNA sequencing and analyses

CD45/Ter119⁻DAPI⁻Calcein-AM⁺ cells were sorted using a yield and then purity sorting mode on BD SORP FACSAria. scRNA-seq was performed on six control and seven Prx1-cre; Mettl3^fl/fl mice with two independent experiments. 10x Genomics Chromium Next GEM Single Cell 3ʹ Reagent Kits v3.1 were used to prepare the library, targeting ~5,000 cells, according to the manufacturer’s instructions. Sequencing was performed at 2 × 100 bp on NovaSeq 6000. Raw reads were mapped to mm10 genome (mm10-2020A provided by 10x Genomics) and counted using CellRanger (v7.0.1) with default parameters. CellBender (v0.2.2) ⁶³ was applied to remove ambient RNA. scRNA-seq data were then analyzed with Seurat (v4.2.0). Cells with nUMI less than 1,000, nGene less than 1,000, and percentage of mitochondrial genes higher than 0.05 were filtered out. Doublets were excluded with DoubletFinder (v2.0.3) ⁶⁴. The above filtering steps were performed on each sample separately. There were 8,154 cells from control and 6,181 cells from Δ/Δ mice passed through the filtering steps. The stromal cells from two independent experiments were merged after removing contaminating hematopoietic and cycling cells from each independent experiment. The following analyses were performed on the merged stromal cells: data transformation and feature selection (SCTransform, regressing out mitochondrial genes and cell cycle status), sample integration (IntegrateData with the 3,000 most variable genes), dimensional reduction (runPCA and runTSNE), unsupervised clustering (FindClusters), and marker identification (FindConservedMarkers). The identity for each cluster was called with reference to a published dataset ⁶⁵. Proportion of cells in each cluster between control and mutant mice were analyzed with scProportionTest package (https://github.com/rpolicastro/scProportionTest) ⁶⁶. Heatmap was generated with Scillus package (https://scillus.netlify.app/). The transcription factor-encoding gene list was downloaded from AnimalTFDB 4.0 ⁶⁷.

In Figure 1, scRNA-seq data were from GSE178951. The data analysis steps were similar as described above. The following parameters were used to filter out low-quality cells: nUMI no less than 4,000, nGene no less than 2,000, log10GenesPerUMI above 0.8, mitochondrial gene percentage less than 7.5%, and ribosomal gene percentage less 40%. GSEA analysis was performed with GSEA software (v4.1.0) ⁶⁸. GO analysis was performed with DAVID online tool.

Lentiviral packaging and Klf2 overexpression

Klf2 overexpression lentiviral vector was constructed by replacing GFP with Klf2 cDNA in FUGW ⁶⁹ (Addgene, Cat#14883) using BamH I and EcoR I. Lentivirus was packaged by transfecting lentiviral vector and packaging vectors (Pax2 and pMD2.G) into HEK293T cells with calcium phosphate. Lentivirus-containing supernatant was collected 48 and 72 hours after transfection. Bone marrow stromal cells were seeded and then infected with lentivirus supernatant 24 later by spinning at 1,400 g for 90 minutes at 25 °C in the presence of 8 μg/ml polybrene. The cells were harvested for RT-qPCR or flow cytometry analysis at 96 hours after the infection.

RT-qPCR

RNA was isolated from cells with Trizol Reagent (ThermoFisher, Cat#15596018) following the manufacturer’s instructions. Reverse transcription was performed with GoScript^™ Reverse Transcriptase (Promega Cat#A5001) following the manufacturer’s instructions. qPCR was performed using GoTaq qPCR Master Mix (Promega, Cat#A6001) on a CFX Connect Real-Time PCR Detection System (Bio-Rad). At least three biological replicates were analyzed. β-Actin was used as an internal control gene unless otherwise stated. The primers used were as follows:

Mettl3, 5’-AAGGAGCCGGCTAAGAAGTC-3’ and

5’-TCACTGGCTTTCATGCACTC-3’;

Klf2, 5’-CCTATCTTGCCGTCCTTTGC-3’ and

5’-CAGACCGTCCAATCCCATG-3’;

Cxcl12, 5’-TGCATCAGTGACGGTAAACCA-3’ and

5’-GTTGTTCTTCAGCCGTGCAA-3’;

Scf, 5’-CCAAGGAGATCTGCGGGAAT-3’ and

5’-TTACCATATCTCGTAGCCAACAATG-3’;

Lepr, 5’-CCTGGGCACAAGGACTGAAT-3’ and

5’-GAGACCATAGCTGCTGGGAC-3’;

Pdgfra, 5’-CCAGTGGCTTTCTGTTTGGC-3’ and

5’-TCGGACCACTGACAGAAAGC-3’;

Adipoq, 5’-CAGTGGATCTGACGACACCAA-3’ and

5’-ATGCCTGCCATCCAACCTG-3’;

Pparg, 5’-TGGGTGAAACTCTGGGAGATTCTC-3’ and

5’-GAGAGGTCCACAGAGCTGATTCC-3’;

Col1a1, 5’-GCATGGCCAAGAAGACATCC-3’ and

5’-CCTCGGGTTTCCACGTCTC-3’;

Runx2, 5’-CCAAGAAGGCACAGACAGAA-3’ and

5’-CGGGACACCTACTCTCATACT-3’;

Ocn, 5’-ACCATCTTTCTGCTCACTCTG-3’ and

5’-GTTCACTACCTTATTGCCCTCC-3’;

Bglap2, 5’-CACCTAGCAGACACCATGAG-3’ and

5’-GTTCACTACCTTATTGCCCTCC-3’;

Alpl, 5’-ACACCTTGACTGTGGTTACTGCTGA-3’ and

5’-CCTTGTAGCCAGGCCCGTTA-3’;

β-Actin, 5’-GCTCTTTTCCAGCCTTCCTT-3’ and

5’-CTTCTGCATCCTGTCAGCAA-3’;

Gapdh, 5’-AACTTTGGCATTGTGGAAGG-3’ and

5’-ATGCAGGGATGATGTTCTGG-3’.

m⁶A ELISA

mRNA was purified from cultured bone marrow stromal cells with Oligo-(dT)25 magnetic Dynabeads (ThermoFisher Scientific, Cat#61011) according to the manufacturer’s instructions and quantified with a spectrophotometer (NanoDrop ND-1000). mRNA m⁶A levels were measured with an m⁶A RNA Methylation Quantification Kit (EpiQuik, Cat#P-9005) following the manufacturer’s instructions. The absorbance was read on a microplate reader (SpectraMax iD3) at 450 nm. At least three biological replicates were analyzed without randomization or blinding. No data were excluded.

Targeted Klf2 mRNA demethylation

dCas13b-Fto and dCas13b-Alkbh5 expression lentiviral vectors were generated by replacing the GFP sequence in the FuGW empty vector with dCas13b-Fto-P2A-GFP or dCas13b-Alkbh5-P2A-GFP, amplified from pcDNA3.1(+)-dCas13b-Fto-P2A-GFP or pcDNA3.1(+)-dCas13b-Alkbh5-P2A-GFP vectors ^47,48. Lentivirus were prepared as described above. Control and Klf2-targeting gRNAs (sequences are provided below) were cloned into the pC0043-PspCas13b crRNA backbone (Addgene, Cat#103854). Cultured bone marrow stromal cells were infected with lentivirus expressing dCas13b-Fto or dCas13b-Alkbh5. GFP⁺CD45/Ter119⁻ non-hematopoietic cells were sorted 72 hours after infection. The sorted cells were cultured for 24 hours and then electroporated with vector containing control gRNA or a pool of 4 Klf2-targeted gRNAs using a 4D-Nucleofector system and a P3 Primary Cell 4D-Nucleofector X kit (Lonza, Cat#V4XP-3032) with the EN-158 program. Cells were harvested for meRIP-qPCR and RT-qPCR analysis 48 hours after electroporation. The following oligos were used for making the gRNA vectors.

Control gRNA: Forward, 5’-caccGGTGATCAACTCGAGGCGGTTATACGCAAGTTATGAATAC-3’; Reverse, 5’-caacGTATTCATAACTTGCGTATAACCGCCTCGAGTTGATCACC-3’. Klf2 gRNA #1: Forward, 5’-caccGTCGACCCAGGCTACATGTGTCGCTTCATGTGCAAGGCCA-3’; Reverse, 5’-caacTGGCCTTGCACATGAAGCGACACATGTAGCCTGGGTCGAC-3’. Klf2 gRNA #2: Forward, 5’-caccATCCCATGGAGAGGATGAAGTCCAACACGTTGTTTAGGTC-3’; Reverse, 5’-caacGACCTAAACAACGTGTTGGACTTCATCCTCTCCATGGGAT-3’. Klf2 gRNA #3: Forward, 5’-caccTAGGTCTTGCCGCAGTTGGTGTAGCTGCAAGTATGTGTGG-3’; Reverse, 5’-caacCCACACATACTTGCAGCTACACCAACTGCGGCAAGACCTA-3’. Klf2 gRNA #4: Forward, 5’-caccGGCTGGCGAAAGTGGCAAAGGACGGCAAGATAGGCTCGCT-3’; Reverse, 5’-caacAGCGAGCCTATCTTGCCGTCCTTTGCCACTTTCGCCAGCC-3’.

Competitive transplantation and limiting dilution assay

B6.SJL-Ptprc^aPepc^b/BoyJ mice from the Jackson Laboratory were used as transplantation recipients. Adult recipient mice were lethally irradiated with a Cesium 137 Irradiator (JL Shepherd and Associates) at 300 rad/min with two doses of 540 rad (total 1080 rad) or with a MultiRad 225 X-ray irradiator (Precision X-Ray, total dose 1050 cGy) with two doses delivered at least two hours apart. For competitive transplantation, 5 × 10⁵ donor and 5 × 10⁵ competitor whole bone marrow cells were transplanted by retro-orbital venous sinus injection. For limiting dilution assay, 10,000, 30,000, or 80,000 bone marrow cells, together with 3 × 10⁵ competitor bone marrow cells, were transplanted. Recipient mice were maintained on antibiotic water (Baytril 0.17 g/L) for 14 days and then switched to regular water. Peripheral blood from recipient mice was analyzed by flow cytometry every four weeks for 16 weeks after transplantation. For limiting dilution assay, recipient mice were considered as multilineage reconstituted if they showed donor-derived myeloid, B, and T cells above background levels in the peripheral blood at 16 weeks after transplantation. Red blood cells were lysed with ammonium chloride potassium lysis buffer before antibody staining. The following antibodies were used: anti-CD45.1 (A20), anti-CD45.2 (104), anti-Gr1 (8C5), anti-Mac1 (M1/70), anti-CD3 (17A2), and anti-B220 (6B2).

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analyses were performed using GraphPad Prism 9 or Excel, unless otherwise stated. For Mendelian ratio analysis, p value was calculated with binomial test. For Kaplan-Meyer survival analysis, p value was calculated by log-rank. For limiting dilution assay, data were analyzed with ELDA software (https://bioinf.wehi.edu.au/software/elda/) ⁵⁵ and p value was calculated with chisq test. For all other comparisons, p value was calculated by two-tailed unpaired t-test, one-way ANOVA followed by Dunnett’s multiple comparison, or two-way ANOVA followed by Bonferroni’s multiple comparisons test as stated in the figure legends. In all figures, dots on bar graphs represent biological replicates and error bars represent standard deviation.

Supplementary Material

Table S1

Table S1. Lists of differentially expressed genes in distinct bone marrow cell clusters between Prx1-cre; Mettl3^fl/fl and control mice, related to Figure 4.

Table S2

Table S2. List of significantly enriched 1,928 genes in anti-m⁶A antibody-bound fraction compared with input, related to Figure 6.

Table S3

Table S3. Lists of significantly enriched GO terms in anti-m⁶A antibody-bound fraction compared with input, related to Figure 6.

Table S4

Table S4. List of DNA oligo sequences used in this paper, related to STAR Methods.

Figure S1. Figure S1. scRNA-seq analysis of perinatal and adult MSCs. Related to Figure 1.

(A) Heatmap showing 12 signature genes for each cluster identified in Figure 1A. PαS, Pdgfrα⁺Sca1⁺ stromal cells; Osteochondro, osteo-chondrocyte progenitors; OLC, osteolineage cells.

(B) Gene set enrichment analysis (GSEA) using all the gene sets from gene ontology (GO) molecular function (MF) ontology. The red and blue dots indicate gene sets enriched in perinatal and adult MSCs, respectively. The dashed line indicates FDR=0.05.

(C and D) The top 20 GO-MF gene sets enriched in perinatal (C) or adult (D) MSCs. The x-axis represents normalized enrichment score (NES). The size and color of each dot represents the number of significantly enriched genes in each gene set and nominal p value, respectively.

(E) GO analysis of the MF terms on the 115 core enriched genes of mRNA binding gene set identified in B and C.

(F) Expression levels of Mettl3 and Mettl14 in perinatal and adult MSCs revealed by scRNA-seq. P values were calculated with two-tailed unpaired t-test.

Figure S2. Figure S2. Prx1-cre recombines in mesenchymal cells of neonatal limb bones and Prx1-cre; Mettl3^fl/fl mice have reduced birth and survival rates. Related to Figure 2.

(A-E) Immunostaining of aggrecan (A), osteopontin (B), perilipin (C), TRAP (D), and CD45 (E) on femur sections from two-week-old Prx1-cre; Rosa26^{LSL-tdTomato/+} mice.

(F) Quantification of tdTomato⁺ cell frequency in indicated cell types in A-E. Data are represented as mean ± SD.

(G) Flow cytometry plots showing tdTomato expression in CD45/Ter119⁻PDGFRα⁺ bone marrow MSCs and PDGFRα expression in tdTomato⁺ stromal cells from the hindlimb bone marrow of two-week-old Prx1-cre; Rosa26^{LSL-tdTomato/+} mice.

(H) Quantification of control and Prx1-cre; Mettl3^fl/fl (Δ/Δ) mice that were alive at P0. P value was calculated with two-tailed binomial test. A total of 150 mice were analyzed.

(I) Kaplan-Meyer survival analysis of control and Δ/Δ mice. P value was calculated by log-rank test (n=99 for control, n=16 for Δ/Δ).

(J) Representative images of control and Δ/Δ mice at P0, P5, and two weeks old. Arrows point to abnormal forelimbs from Δ/Δ mice.

(K) Representative images of femurs and tibias from control and Δ/Δ mice at P0, P5, and two weeks old.

(L-M) Representative images of H&E staining on the femur sections from control and Δ/Δ mice at P0 (L) and two weeks old (M). Black boxes in low-mag images on the left indicate the magnified regions shown on the right.

Figure S3. Figure S3. Prx1-cre; Mettl3^fl/fl mice have decreased HSC frequency and number in the bone marrow while Ocn-cre; Mettl3^fl/fl and Lepr-cre; Mettl3^fl/− mice have normal hematopoiesis in the bone marrow. Related to Figures 2 and 3.

(A) Representative flow cytometry plots showing HSC staining on bone marrow cells from control and Prx1-cre; Mettl3^fl/fl (Δ/Δ) hindlimbs.

(B) HSC frequencies in the peripheral blood from two-week-old control and Δ/Δ mice.

(C) HSC distribution in the livers, spleens, vertebral, and limb bones of 4 limbs from two-week-old control and Δ/Δ mice. Two-way ANOVA with Bonferroni’s multiple comparison tests were used to calculate the p values.

(D) Numbers of recipients and responders for indicated conditions of limiting dilution assay. 10,000, 30,000, or 80,000 whole bone marrow cells from hindlimbs of control or Δ/Δ mice, together with 300,000 competitor cells, were transplanted into lethally-irradiated recipients. Recipients were called responders if they showed donor-derived myeloid, B, and T cells in the peripheral blood at 16 weeks after transplantation. Samples were from independent recipients of two independent donor pairs from two experiments.

(E) Frequencies of functional HSCs in the hindlimb bone marrow from control and Δ/Δ mice revealed by limiting dilution transplantation assay. Data were analyzed with ELDA ⁵⁵.

(F-H) Immunostaining of osteopontin (F), aggrecan (G), and perilipin (H) on femur sections from two-week-old Ocn-cre; Rosa26^{LSL-tdTomato/+} mice.

(I) Quantification of tdTomato⁺ cell frequencies in indicated cell types in F-H.

(J) Schematic diagram showing the breeding strategy to generate Mettl3^fl/− and Lepr-cre; Mettl3^fl/− mice.

(K-N) Bone marrow cellularity (K), CD45/Ter119⁻PDGFRα⁺ MSC frequency (L), LSK frequency (M), and HSC frequency (N) in two hindlimbs of adult Mettl3^fl/− and Lepr-cre; Mettl3^fl/− mice.

Data are presented as mean ± SD. P values were calculated with two-tailed unpaired t-test unless otherwise noted.

Figure S4. Figure S4. scRNA-seq analyses of bone marrow stromal cells from control and Prx1-cre; Mettl3^fl/fl mice. Related to Figure 4.

(A) Flow cytometry plots showing the gating strategy for sorting DAPI⁻Calcein-AM⁺CD45/Ter119⁻ bone marrow cells from hindlimbs of control and Prx1-cre; Mettl3^fl/fl (Δ/Δ) mice for scRNA-seq analyses.

(B) Heatmap showing ten signature genes for each cluster in Figure 4B.

(C) t-SNE plots showing the expression of indicated genes in the bone marrow stroma. Cells from control and Δ/Δ mice were aggregated. A total of 5,726 cells, including 1,870 cells from control and 3,856 cells from Δ/Δ mice, are shown.

(D) Violin plots showing the expression levels of Cxcl12, Scf, and Lepr in each cell cluster identified in control and Δ/Δ mice.

Figure S5. Figure S5. Prx1-cre; Mettl3^fl/fl mice have more osteolineage cells and less MSCs in the bone marrow. Related to Figure 5.

(A-N) Immunostaining of osteopontin (A), aggrecan (C), VE-cadherin (E), TRAP (G), and CD45 (I) on femur sections and Nile Red (K), and LipidTox (M) on whole-mount femurs from two-week-old control and Prx1-cre; Mettl3^fl/fl (Δ/Δ) mice. The quantification results of relative positively stained areas for each marker are showed in B, D, F, H, J, L, and N.

(O and P) Representative confocal images of femur sections from P0 Cxcl12^DsRed/+ and Prx1-cre; Mettl3^fl/fl; Cxcl12^DsRed/+ mice. The quantification results of Cxcl12-DsRed⁺ areas in the bone marrow are showed in P.

Data are represented as mean ± SD. P values were calculated with two-tailed unpaired t-test.

Figure S6. Figure S6. Klf2 is a target of m⁶A modification in neonatal bone marrow MSCs. Related to Figure 6.

(A) PCA plot of indicated fractions revealed by meRIP-seq. A total of 9,231 genes with mean abundance in at least one fraction was larger than the median value of all detectable genes were used to generate the plot.

(B) Volcano plot showing the difference of mRNA abundance between anti-m⁶A-bound and input fractions. The dashed vertical and horizontal lines indicate the filtering criteria (|log₂FoldChange| > 2 and adjusted p value < 0.05). 1,928 genes are enriched in the anti-m⁶A-bound fraction compared with input. Klf2 mRNA is enriched in the anti-m⁶A-bound fraction. Genes with mean abundance in at least one fraction was larger than the median value of all detectable genes were analyzed to reduce uncertainty.

(C) GO analysis of the 1,928 m⁶A target genes identified in B.

(D) Schematic diagrams showing bone marrow stromal cells were cultured from hindlimbs of two-week-old control and Prx1-cre; Mettl3^fl/fl (Δ/Δ) mice.

(E) Relative Klf2 expression levels at the indicated time points after actinomycin-D treatment in cultured bone marrow stromal cells from two-week-old control and Δ/Δ mice. Klf2 mRNA levels were normalized to Gapdh in DMSO-treated control or Mettl3-deficient cells.

(F) Schematic diagrams depicting bone marrow stromal cells cultured from the hindlimbs of two-week-old mice were transfected with dCas13b-Fto or dCas13b-Alkbh5, then with control or Klf2 gRNAs. Four gRNAs targeting different regions of Klf2 mRNA were combined.

(G and H) m⁶A levels on Klf2 revealed by meRIP-qPCR (G) and Klf2 mRNA levels revealed by RT-qPCR (H) in treated groups as described in F.

(I) Expression levels of indicated genes in cultured bone marrow stromal cells isolated from hindlimbs of two-week-old control and Δ/Δ mice as shown in D. Hematopoietic niche-associated genes, adipolineage markers, and osteolineage markers are shadowed in red, blue, and purple, respectively.

Data are presented as mean ± SD. P values were calculated with two-tailed unpaired t-test.

Figure S7. Figure S7. Klf2 deletion rescues hematopoietic niche formation defects of Prx1-cre; Mettl3^fl/fl mice. Related to Figure 7.

(A) Detailed genotypes of mice in Control, Klf2-del, Mettl3-del, and Double-del groups.

(B) Quantification of Control, Klf2-del, Mettl3-del, and Double-del mice that were alive at P0. P value was calculated with one-tailed binomial test. A total of 333 mice were analyzed.

(C-G) Representative immunostaining images of osteopontin (C), aggrecan (D), CD45 and TRAP (E), VE-cadherin (F), and Nile Red (G) on the femur sections from two-week-old Control, Klf2-del, Mettl3-del, and Double-del mice.

(H) Quantification of the immunostaining results in C-G. Data were normalized to the Control group.

(I) Representative flow cytometry plots showing the frequencies of CD45/Ter119⁻PDGFRα⁺ cells in the hindlimb bone marrow of two-week-old Control, Klf2-del, Mettl3-del, and Double-del mice.

(J) Frequencies of CD3⁺ T cells, CD41⁺ megakaryocytic cells, Mac1⁺Gr1⁺ myeloid cells, and Ter119⁺ erythrocytes in the hindlimb bone marrow of two-week-old Control, Klf2-del, Mettl3-del, and Double-del mice.

(K) Representative flow cytometry plots showing the frequencies of HSCs in the hindlimb bone marrow of two-week-old Control, Klf2-del, Mettl3-del, and Double-del mice.

Data are represented as mean ± SD. P values were calculated with two-tailed unpaired t-test.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
anti-CD2 (clone RM2-5)	BioLegend	Cat#100105; RRID:AB_312652
anti-CD3 (clone 17A2)	BioLegend	Cat#100204; RRID:AB_312660
anti-CD5 (clone 53-7.3)	BioLegend	Cat#100605; RRID:AB_312734
anti-CD8a (clone 53-6.7)	BioLegend	Cat#100705; RRID:AB_312744
anti-Ter119 (cloneTer119)	BioLegend	Cat#116206; RRID:AB_313707
anti-B220 (clone 6B2)	BioLegend	Cat#103205; RRID:AB_312991
anti-Gr1 (clone 8C5)	BioLegend	Cat#108406; RRID:AB_313370
anti-Sca-1 (clone E13-161.7)	BioLegend	Cat#122514; RRID:AB_756199
anti-cKit (clone 2B8)	BioLegend	Cat#105826; RRID:AB_1626280
anti-CD150 (clone TC15-12F12.2)	BioLegend	Cat#115904; RRID:AB_313682
anti-CD48 (clone HM48-1)	BioLegend	Cat#103412; RRID:AB_571996
anti-CD41 (clone MWReg30)	BioLegend	Cat#133906; RRID:AB_2129745
anti-Mac1 (clone M1/70)	BioLegend	Cat#101212; RRID:AB_312795
anti-CD45 (clone 30F-11)	BioLegend	Cat#103106; RRID:AB_312971
anti-CD140α (clone APA5)	BioLegend	Cat#135910; RRID:AB_2043973
anti-CD31(clone 390)	BioLegend	Cat#102409; RRID:AB_312904
anti-CXCL12 (clone 79018)	R&D Systems	Cat#IC350F; RRID:AB_1964551
anti-METTL3	BosterBio	Cat#A01758-1
anti-aggrecan	Millipore	Cat#AB1031; RRID:AB_90460
anti-osteopontin	R&D Systems	Cat#AF808; RRID:AB_2194992
anti-perilipin (clone D1D8)	CST	Cat#9349S; RRID:AB_10829911
anti-VE-cadherin	R&D Systems	Cat#BAF1002; RRID:AB_2260340
anti-TRAP	Abcam	Cat#ab185716
anti-m6A	Synaptic Systems	Cat#202003; RRID:AB_2279214
anti-CD45.1 (clone A20)	BioLegend	Cat#110716; RRID:AB_313504
anti-CD45.2 (clone 104)	BioLegend	Cat#109806; RRID:AB_313442
Biological samples
Fetal bovine serum	Gemini	Cat#100-106
Chemicals, peptides, and recombinant proteins
Collagenase IV	Worthington Biochemical	Cat#LS004186
DNase I	MilliporeSigma	Cat#260913
Calcein-AM	BioLegend	Cat#425201
Liberase	Roche	Cat#5401119001
OCT solution	Fisher Scientific	Cat#23-730-571
Nile Red	MilliporeSigma	Cat#N3013
LipidTox	Invitrogen	Cat#H34476
Protein A/G magnetic beads	MilliporeSigma	Cat#16-663X
N6-methyladenosine 5-monophosphate sodium salt	MilliporeSigma	Cat#M2780
Actinomycin D	MilliporeSigma	Cat#A9415
Trizol Reagent	Thermo Fisher	Cat#15596018
GoScript™ Reverse Transcriptase	Promega	Cat#A5001
GoTaq qPCR Master Mix	Promega	Cat#A6001
αMEM	Gibco	Cat#12571063
DMEM	Corning	Cat#10-017-CM
Antimicrobial Injection Solution (Baytril)	Elanco	Cat#Baytril 100
Polybrene	MilliporeSigma	Cat#TR-1003-G
L-glutamine	Corning	Cat#25-015-CI
Pen Strep	Gibco	Cat#15140-122
Sodium Pyruvate	Gibco	Cat#11360-070
Trypsin-EDTA Solution	MilliporeSigma	Cat#59417C
Critical commercial assays
Chromium Single Cell 3’ Reagent Kit (V3)	10x Genomics	Cat#PN-1000269
eBioscience Transcription Factor Staining kit	Thermo Fisher	Cat#00-5523-00
Dynabeads™ mRNA DIRECT™ Purification Kit	Thermo Fisher	Cat#61011
NextSeq 500/550 High Output Kit	Illumina	Cat#20024907
m6A RNA Methylation Quantification Kit	EpiQuik	Cat#P-9005
P3 Primary Cell 4D-Nucleofector X kit	Lonza	Cat#V4XP-3032
Deposited data
Raw and analyzed data	This paper	GEO: GSE217057
Raw and analyzed data	Hall et al. 15	GEO: GSE178951
Experimental models: Cell lines
HEK293T	ATCC	Cat#CRL-3216
Experimental models: Organisms/strains
Prx1-cre	Logan et al. 34	JAX:005584; RRID:IMSR_JAX:005584
Lepr-cre	DeFalco et al. 56	JAX:032457; RRID:IMSR_JAX:032457
Mettl3 fl/fl	Geula et al. 51	N/A
Cxcl12 DsRed	Ding et al. 9	JAX:022458; RRID:IMSR_JAX:022458
Klf2 gfp	Weinreich et al. 45	JAX:036331; RRID:IMSR_JAX:036331
Klf2 fl	Weinreich et al. 45	N/A
Ocn-cre	Zhang et al. 37	JAX:019509; RRID:IMSR_JAX:019509
B6.SJL-PtprcaPepcb/BoyJ	The Jackson Laboratory	JAX:002014; RRID:IMSR_JAX:002014
Rosa26 LSL-ZsGreen	Madisen et al. 57	JAX:007906; RRID:IMSR_JAX:007906
Rosa26 LSL-tdTomato	Madisen et al. 57	JAX:007909; RRID:IMSR_JAX:007909
Oligonucleotides
qPCR primers	This paper	Table S4
gRNA oligos	This paper	Table S4
Recombinant DNA
Fugw	Lois et al. 69	Addgene 14883
Fugw-Klf2	This paper	N/A
Fugw-dCas13b-Fto-P2A-GFP	This paper	N/A
Fugw-dCas13b-Alkbh5-P2A-GFP	This paper	N/A
pC0043-PspCas13b crRNA-control	This paper	N/A
pC0043-PspCas13b crRNA-Klf2_#1	This paper	N/A
pC0043-PspCas13b crRNA-Klf2_#2	This paper	N/A
pC0043-PspCas13b crRNA-Klf2_#3	This paper	N/A
pC0043-PspCas13b crRNA-Klf2_#4	This paper	N/A
Software and algorithms
FlowJo	FlowJo LLC	https://www.flowjo.com/solutions/flowjo/downloads
GraphPad Prism 9.0	GraphPad Prism 9.0	https://www.graphpad.com
Fiji	Fiji	https://fiji.sc/
RSubread	Liao et al. 60	https://bioconductor.org/packages/release/bioc/html/Rsubread.html
Picard	Broad Institute	http://broadinstitute.github.io/picard
edgeR	Robinson et al. 61	https://bioconductor.org/packages/release/bioc/html/edgeR.html
CellRanger	10x Genomics	https://support.10xgenomics.com/single-cell-gene-expression/software/downloads/latest
CellBender	Fleming et al. 62	https://github.com/broadinstitute/CellBender
Seurat	Satija Lab	https://satijalab.org/seurat/
DoubletFinder	McGinnis et al. 63	https://github.com/chris-mcginnis-ucsf/DoubletFinder
scProportionTest	Miller et al. 65	https://github.com/rpolicastro/scProportionTest
Scillus	N/A	https://scillus.netlify.app/
ELDA	Hu et al. 55	https://bioinf.wehi.edu.au/software/elda/
R	the R Foundation	https://www.r-project.org/
GSEA	Subramanian et al. 67	https://www.gsea-msigdb.org/gsea/index.jsp

Highlights of the paper:

Perinatal bone marrow MSCs preferentially express m⁶A-associated genes relative to adults

Deletion of Mettl3 from developing MSCs leads to a severe HSC niche formation defect

Deletion of Mettl3 from MSCs postnatally does not affect HSC niche maintenance

Mettl3 controls HSC niche formation by regulating Klf2 expression in MSCs